CLN: Simplify gathering of results in aggregate

[rhshadrach]

• closes #xxxx
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

Reduces the number of paths when collecting results in aggregation.aggregate; one where we are gathering NDFrames using concat and the other where we are gathering scalars using Series.

The reason for the one test change is as follows. In one case, we previously gathered results using DataFrame. When this occurs and the indexes are not all equal, DataFrame will sort the index whereas concat will have the index in order of appearance. For example with

df = DataFrame(
    {
        'A': pd.Series([1, 2], index=['b', 'a']),
        'B': pd.Series([3, 4], index=['c', 'a'])
    }
)

gives

     A    B
a  2.0  4.0
b  1.0  NaN
c  NaN  3.0

whereas using concat instead of DataFrame on the first line with axis=1 gives:

     A    B
b  1.0  NaN
a  2.0  4.0
c  NaN  3.0

If in this example you replace the 2nd index with ['b', 'a'] (so that they are equal), then both Dataframe and concat will produce the same result with index ['b', 'a']. If on the other hand you replace the 2nd index with ['a', 'b'], then DataFrame will result in index ['a', 'b'] whereas concat will result in index ['b', 'a'].

[jreback]

wow a lot simpler. pls merge master and ping on green.

[jreback]

can you add a ABCNDFrame in pandas.core.dtypes.generic

[jreback]

can you move this comment below the AttributeError (as confusing where it is now)

[jreback]

where does the AttributeError come from?

can you move things outside the try/except (or use an else)?

[jreback]

lgtm merge on green

[jreback]

thanks @rhshadrach